Textify

Third-party software

The application bundle includes license texts in textify/licenses, the whisper.cpp license in textify/WHISPER-LICENSE, and Electron/Chromium notices provided by Electron. Node packages retain their own notices in the archive.

The model browser includes publisher logos solely to identify the source of each model. These logos and names belong to their respective owners; their appearance does not imply endorsement of Textify. The bundled images came from the publishers’ public profiles or site:

• OpenAI: https://huggingface.co/openai
• NVIDIA: https://nvidianews.nvidia.com/multimedia/corporate/nvidia-logos
• Qwen: https://github.com/QwenLM

• NetEase Youdao: https://dict.youdao.com/home

• Electron and Chromium: https://github.com/electron/electron
• React and React DOM, MIT: https://github.com/facebook/react
• whisper.cpp, MIT, pinned to a8d002cfd879315632a579e73f0148d06959de36: https://github.com/ggml-org/whisper.cpp
• OpenAI Whisper and small.en, large-v2, large-v3, large-v3-turbo models, MIT: https://github.com/openai/whisper
• Silero VAD v5.1.2 speech detector, MIT, bundled as textify/ggml-silero-v5.1.2.bin from ggml-org/whisper-vad revision 9ffd54a1e1ee413ddf265af9913beaf518d1639b; the license is included in textify/licenses/Silero-VAD.txt: https://github.com/snakers4/silero-vad
• uiohook-napi, MIT: https://github.com/SnosMe/uiohook-napi
• libuiohook, LGPL-3.0-or-later, copyright Alexander Barker: https://github.com/kwhat/libuiohook
• dbus-next (Delta Chat fork), MIT: https://github.com/deltachat/node-dbus-next
• electron-updater and builder-util-runtime, MIT: https://github.com/electron-userland/electron-builder. The Mac updater also bundles debug, fs-extra, graceful-fs, has-flag, js-yaml, jsonfile, lazy-val, lodash.escaperegexp, lodash.isequal, ms, sax, semver, supports-color and universalify under MIT, ISC or BlueOak-1.0.0 terms. lazy-val declares MIT and publishes no separate license file.

The installed uiohook-napi source, its included libuiohook source, and build description are provided in textify/native-source/uiohook-napi. Its native module is external to the application archive under app.asar.unpacked and can be replaced with a rebuilt module. Use its included binding.gyp with node-gyp for the target platform. Changes to a macOS bundle require re-signing it.

Whisper weights are downloaded or imported separately, verified against the signed catalog, and are not included in the application package.

Windows/Linux builds use Vulkan-Headers and Vulkan-Hpp (Khronos Group), Apache-2.0; the license is included in textify/licenses/Vulkan.txt. Vulkan drivers and the system Vulkan loader are external dependencies, not bundled. The Vulkan SDK/glslc is used at build time; end users do not need the SDK. The pinned whisper source is modified by native/require-gpu.mjs to refuse CPU graph execution, require GPU weights/backends and reject software Vulkan devices. The one exception is the Silero speech detector, which runs on the CPU by design.

All builds also contain isolated, statically linked GPU workers, using Metal on macOS and Vulkan on Windows and Linux:

• transcribe.cpp 0.1.3, MIT, commit 5a5a49664a8ea1f0e5b3be1dfc544730d1b62561: https://github.com/handy-computer/transcribe.cpp
• audio.cpp, Apache-2.0, commit 9ba884179826c3b33dd305185b5f94c79175a03d: https://github.com/0xShug0/audio.cpp

The checksum-pinned source archives include their exact ggml and tokenizer sources. native/prepare-extra.mjs adds CPU graph refusal, rejects software Vulkan devices and moves Parakeet TDT’s predictor/joint graphs to the GPU. The audio.cpp C ABI is linked statically into its worker so its ggml symbols do not collide with other runtimes. Bundled transcribe.cpp.txt and audio.cpp.txt include dependency notices. Qwen3-ASR, Parakeet TDT and Confucius4-R2T2 weights are downloaded or imported separately and remain subject to the licenses shown in the signed catalog. Confucius uses the NetEase Youdao Model Use License Agreement; it is not MIT. The original BF16 checkpoint is pinned to NetEase Youdao’s Hugging Face revision 185ce639118ad1362d049ca0d8ed04b6ec5cd6c9. Its original weights are not converted or quantized by Textify. The existing bundled Confucius4-R2T2.txt contains the publisher’s model terms. audio.cpp is built with its model specifications embedded for offline safetensors-directory loading.

This site is open source. Improve this page.